A Research Of Computer-Aided Drug Design On The Basis Of Aurora Kinase

Aurora kinase A is a widely recognized anti-cancer target, it is important to design new hit inhibitors of Aurora kinase A. Using different computer aided drug design methods, through studying of relationship between the known structure-activity of Aurora kinase A inhibitors, including building the classification models and quantitative prediction models, with the virtual screening on the basis of the small molecule ligand compounds, several hit compounds with highly active potential for inhibition of Aurora kinase A has been chosen out from small molecules databases. This thesis can mainly be summarized in the following three parts:（1） The study of classification models for Aurora kinase A inhibitors based on Self-organizing Map （SOM） and Support Vector Machine （SVM） methods. It was selected29ADRIAN A. Code descriptors for1463Aurora kinase A inhibitors with the known structures and activities, and the classification models including ModelAl and ModelA2, were established by the methods of SOM and SVM, respectively. The accuracies for ModelA1 on the training set and the test set were91.19%and86.10%, while the prediction accuracies for ModelA2on the training set and the test set were94.13%and86.11%. In addition, there was an analysis on the relationship between biology activity and substructure of inhibitors by ECFP₄fingerprint methods.（2） Development of quantitative prediction models for the bioactivity of Aurora kinase A inhibitors based on MLR and SVM methods. According to three different kinase assays for activity, the inhibitors were divided into three subsets including356/302/279compounds. Each subset was divided into different training and test sets by random and SOM methods, and altogether12prediction models were established by Multilinear Regression and Support Vector Machine methods, respectively. The correlation coefficient R values of all models on test set are not less than0.77. All the models were examined through Y-randomization method, which showed favorable prediction ability.（3） The study of virtual screening based on Aurora kinase A inhibitors as ligands. The software ROCs and EON were used to explore how to establish best query in different ways. The VX680was selected as the best query finally. A number of500optimal compounds which most similar with the query was selected from nearly five million small molecules with virtual screening.500compounds were considered as high active through the screening with the optimal classification model, and there found23compounds whose predicted bioactivity values are less than10nM. Through the research on classification and quantitative prediction based on biological activity of Aurora kinase A inhibitors, several computational model were established. In the study of virtual screening by3D molecular of Aurora kinase A inhibitors about similar of their shape and electrostatic, some hit compounds with potential inhibitory activity were found.